After finding some discussion in the kernel list about SMP problems and
lilo, I tried booting this Netfinity 7000 from a floppy (no lilo).
Success for the first time - 2.2.4 SMP came up with 2 GB of memory
available, and it's been stable through a number of kernel builds and
various benchmarks.
So, I guess lilo is corrupting some critical area for SMP operations.
During boot there is this warning:
WARNING: MP table in the EBDA can be UNSAFE, contact
[EMAIL PROTECTED] if you experience SMP problems!
I built the newest lilo with the -DLARGE_EBDA switch and re-ran lilo, but
still run into the kernel panic. Perhaps the boot loader needs to be
moved even further down on this machine.
Don Holmgren
Fermilab
On Thu, 8 Apr 1999, Don Holmgren wrote:
>
> I'd sent mail to the list last week detailing difficulties with SMP on a
> system based on the Intel AD450NX motherboard. I'm sending mail again, as I
> have some further information which might give some clues. Also, I'm hoping
> to get some advice back this time :) .
>
> The system is an IBM Netfinity 7000 10M on loan from IBM. It has 4
> processors (450 MHz Pentium Xeon II), an I/O board with dual Adaptec 7895
> controllers, two fully populated memory boards (8 banks, 2GB total), and AMD
> pcnet32 ethernet.
>
> Under the various mono-kernels (2.0 - 2.2), the system can use all of the
> available memory (either ~1GB, or under 2.2.x ~2GB after modification of the
> PAGE_OFFSET value to change the physical/virtual split from 1:3 to 2:2).
>
> The trouble comes under the SMP kernels. If I use more than 96M (controlled
> via "mem=xx" bootparam's), the system panics during boot. Under the 2.2
> kernel it happens shortly after mounting the root filesystem, with a "Kernel
> panic: NO init found...". Further, the Adaptec controllers on IBM's I/O
> board work fine under mono-kernels, but fail during the boot on SMP kernels
> with the familiar infinite timeout/retry loop. I've put in a Buslogic card,
> which works OK under both mono and SMP kernels (well, OK under 2.0.36 and
> 2.2.x, but not OK under 2.1.13x SMP).
>
> I suspected some sort of memory misconfiguration, so last night I started
> playing with the various controls accessible via the BIOS - write-through vs
> write-back on the Xeons, disabling banks, etc... Sure enough, the system is
> stable if I disable all the banks but one, leaving 256 MBytes of memory.
> With only one bank AFAIK the system can't/won't enable address bit permuting
> or card-to-card memory interleaving.
>
> Thanks in advance. I've appended a dmesg output below.
>
> Don Holmgren
> Fermilab
>
-
Linux SMP list: FIRST see FAQ at http://www.irisa.fr/prive/mentre/smp-faq/
To Unsubscribe: send "unsubscribe linux-smp" to [EMAIL PROTECTED]
